AES Closed Apps June 8 With "No Leetcode." The Bake-Off Is the New Screen.
The June 2026 HN "Who is Hiring" thread normalized paid bake-offs and AI-built portfolios. Here is what that breaks in sourcing, and how to fix it.
The June 2026 "Ask HN: Who is hiring?" thread (item 48357725, posted June 1) reads differently than last year's. Companies aren't asking for a resume and a phone screen. They're posting flat fees, one-week parallel trials, and a single sentence: show us your best AI-built work. If you're sourcing engineers in 2026, the finalist round just moved, and your top-of-funnel has to move with it.
The clearest example sits a few posts down. AES (Associated Environmental Systems, the team behind their FSM platform) wrote it plainly: "Apps close June 8. No leetcode. No whiteboard. Show us your best AI-built work, a PR, a deployed app, or a repo." Applications closed last week. No coding screen. No take-home. The submission was the take-home.
This isn't a one-off. It's the structural shift that techinterview.org's May 2 "30-year arc" essay forecasted, arriving faster than anyone expected.
What actually changed in the June thread
techinterview.org's frame is useful: thirty years of interviewing has moved from Microsoft brainteasers in the 1990s, to the Google whiteboard in the 2000s, to LeetCode standardization in the 2010s, and now into a 2026 fragmentation where coding is still central but the format is contested. The essay is careful: LeetCode isn't dead. Most FAANG and tier-2 coding rounds still use one or two LeetCode-style problems. A generation of candidates is visibly questioning whether it predicts impact, and senior engineers are openly skeptical, but the format hasn't been displaced at the top of the market.
What has been displaced is the unpaid take-home at small teams. Karat's 2026 interview report, drawn from 400 engineering leaders across the US, India, and China, puts it bluntly: take-homes only look at final output, interviewers lack visibility into process, and companies relying on them in 2026 "will struggle to make confident hiring decisions." AI made the artifact cheap. The signal degraded. Something had to fill the gap.
That something is the paid bake-off. A flat fee, a short window, two or three candidates working in parallel on the actual codebase, and a hiring decision at the end. The June HN thread is the first public timestamp of it going mainstream at startup scale.
Why the AES posting matters more than the dollar figure
AES didn't just drop "no leetcode" as a vibe. They named the bar: "the bar is NOT raw coding ability, the team uses Claude Code for all code generation, and the bar is the ability to direct an AI at production engineering and catch it when it's wrong, schema literacy, diff-reading, end-to-end ownership."
Read that twice. Three of the four signals (schema literacy, diff-reading, end-to-end ownership) do not appear on a resume. They do not appear in a LinkedIn headline. They appear in commit history, in PR review threads, in deployment logs, in the blog post where someone explained why their migration broke. The stack they listed (Next.js 14 App Router, RSC, Server Actions, TypeScript strict, Prisma, Postgres, Tailwind, shadcn/ui, NextAuth, Twilio, Anthropic SDK) is the shape of the modern AI-first build. Sourcing against that stack from a resume database is almost useless. Sourcing against it from shipped repos is the entire game.
The economics of a $5K week
The editor's hook references ReadiFinancial offering $5K flat for a roughly one-week parallel trial with two or three candidates. I'd encourage you to verify the exact post text on hnhiring.com/june-2026 before quoting numbers in your own writing. The structural point holds either way: three parallel candidates at $5K each is $15K of cash burn per requisition before you've made an offer.
That math only works if the top of the funnel is already 95% qualified on shipping evidence. You cannot bake-off your way through a noisy shortlist. The unpaid take-home era let you cast wide because the only cost was the candidate's weekend. The paid era forces precision.
This is the real cost-of-being-wrong shift. When the assessment is free, hiring managers tolerate noise. When the assessment costs $15K per req and a week of an engineer's review time, every candidate you advance needs to already look like a hire on paper, or rather, on GitHub.
The unpaid take-home is dead. Small teams replaced it with paid trials. Your sourcing pipeline did not get the memo.
Sourcing has to invert: artifacts first, resume last
The traditional sourcing primitive is "five years of TypeScript, Next.js experience, ideally fintech." That query returns a LinkedIn list. The 2026 primitive is "engineers who shipped a Next.js plus Anthropic SDK app in the last 90 days, with public PR history showing they review their own AI output." That query returns a different list, and the overlap with the first one is smaller than you'd think.
This is exactly the wedge Refolk is built for. You describe the person in plain English ("solo or two-person team, shipped a production app on Next.js 14 with the Anthropic SDK in the last quarter, has public PRs where they pushed back on agent output"), and you get a ranked shortlist across GitHub, LinkedIn, and the open web. The query is the spec. The output is the bake-off candidates.
The "AI-direction" signal is invisible to ATS
Karat's data on the 34% productivity boost is the macro justification, but it also explains why traditional sourcing breaks. AI lifts strong engineers more than weak ones. "Strong" in 2026 means "can direct an agent at a production problem and catch it when it's wrong." None of those words appear on a resume. They appear in:
- Commit messages that reference the prompt that produced the diff
- PR reviews where the author explains why they rejected the agent's first pass
- Blog posts, YouTube videos, and X threads where the engineer argues about agent failure modes
- GitHub issues filed against tools like Claude Code, Cursor, or the Anthropic SDK itself
Tenki's June DevRel posting in the same thread is the mirror image of this from the marketing side: they explicitly want "real engineering experience plus a public technical audience, GitHub with stars, a blog running for a couple years, a YouTube channel engineers watch, or X threads people argue with." That posting tells you what the bake-off buyers are also screening for, even when they don't say it out loud. Public output is the new resume because it's the only thing that survives the AI-output deflation.
Who actually takes a $5K bake-off
Look at who's supplying the candidates. The Pragmatic Engineer's 2026 state-of-the-market read, plus broader hiring data, points at a clear pattern: junior roles still exist with on-ramps, senior and staff still move for deep specialists, but the mid-level tier (three to seven years, generalist, no AI portfolio) is reporting the longest searches, the most ghosting, and the steepest comp cuts.
Those engineers are the natural supply for a paid week of work. They have the skills, they don't have the artifact, and a flat fee plus a real codebase is a faster path to an offer than another six-round loop with a take-home that may or may not be reviewed.
The comments under the June thread name every failure mode: fake jobs, scammers, dodgy take-homes containing malicious packages, day-of cancellations, roles cancelled mid-process, video recording requests, zero feedback. Posting a flat fee in the HN thread itself is a trust signal. It's recruiting funnel marketing as much as it's assessment. Founders who get this are using the fee to skip the loop everyone hates and pull a smaller, better-qualified group directly into work.
What this means for the next 90 days
Three concrete shifts if you're sourcing into this market:
1. Pre-qualify on shipped artifacts, not job titles. If your shortlist for an AI-first role doesn't include a link to a deployed app or a recent PR for every candidate, you're going to burn $15K on bake-offs that should never have started. Refolk's plain-English queries are designed for this: "find me engineers in the EU who have shipped an LLM-powered B2B SaaS feature in the last six months and have public commits to the relevant SDK."
2. Read the founder's posting like a spec. When AES names Claude Code, Next.js 14 RSC, Prisma, and the Anthropic SDK, that's not a wish list. That's the codebase the bake-off will happen in. Source for the exact stack overlap, not the keyword.
3. Source the candidate-side trust signal. Engineers in 2026 are screening you on whether the loop is paid, short, and feedback-bearing. If you can name three reference customers of your assessment process (former bake-off candidates who'll vouch for the experience), you'll out-recruit teams running 7-stage funnels even at lower comp.
The June thread is the first public confirmation that the assessment market split. FAANG kept the live coding interview because they have the volume to justify it. Startups paid up for the bake-off because they don't. The sourcing layer that fed both for a decade (keyword resume matching) doesn't survive either path. Plain-English search against shipped work does, and that's where the next two quarters of hiring will be won.
FAQ
Is LeetCode actually dead in 2026?
No. techinterview.org's May 2 essay is explicit: most FAANG and tier-2 coding rounds still use one or two LeetCode-style problems, and the format is dominant if contested. What's dying is the unpaid take-home at startups, which paid bake-offs are replacing because AI made take-home output too cheap to read as signal.
How do I source for "AI direction" if it's not on resumes?
You source against artifacts and public output. Look at commit history showing PR reviews of AI-generated code, blog posts and YouTube videos discussing agent failure modes, and GitHub issues filed against tools like Claude Code or the Anthropic SDK. This is the kind of query Refolk is built to run in plain English across GitHub, LinkedIn, and the open web at once.
Are paid bake-offs only for senior roles?
The June thread suggests the opposite. The natural supply is mid-level engineers (three to seven years, generalist, no AI portfolio) who are reporting the worst search outcomes in 2026 and need a flat fee plus real code to convert into an offer. Senior specialists are still moving through targeted outreach, not bake-offs.
What should I verify before quoting the ReadiFinancial $5K number?
Open hnhiring.com/june-2026 or news.ycombinator.com/item?id=48357725 and pull the exact post text. The structural shift (paid trials replacing unpaid take-homes) is well documented across the thread and Karat's 2026 data, but specific dollar figures and parallel-candidate counts should always be sourced to the original posting before you cite them in outreach or internal memos.